3 * This file is part of LyX, the document processor.
4 * Licence details can be found in the file COPYING.
8 * Full author contact details are available in file CREDITS.
13 #include "support/docstream.h"
14 #include "support/unicode.h"
24 using lyx::ucs4_codeset;
28 // We use C IO throughout this file, because the facets might be used with
29 // lyxerr in the future.
32 /// codecvt facet for conversion of UCS4 (internal representation) to UTF8
33 /// (external representation) or vice versa
34 class iconv_codecvt_facet : public codecvt<lyx::char_type, char, mbstate_t>
36 typedef codecvt<lyx::char_type, char, mbstate_t> base;
38 /// Constructor. You have to specify with \p inout whether you want
39 /// to use this facet only for input, only for output or for both.
40 explicit iconv_codecvt_facet(string const & encoding = "UTF-8",
41 ios_base::openmode inout = ios_base::in | ios_base::out,
43 : base(refs), encoding_(encoding)
45 if (inout & ios_base::in) {
46 in_cd_ = iconv_open(ucs4_codeset, encoding.c_str());
47 if (in_cd_ == (iconv_t)(-1)) {
48 fprintf(stderr, "Error %d returned from iconv_open(in_cd_): %s\n",
49 errno, strerror(errno));
51 throw lyx::iconv_codecvt_facet_exception();
54 in_cd_ = (iconv_t)(-1);
55 if (inout & ios_base::out) {
56 out_cd_ = iconv_open(encoding.c_str(), ucs4_codeset);
57 if (out_cd_ == (iconv_t)(-1)) {
58 fprintf(stderr, "Error %d returned from iconv_open(out_cd_): %s\n",
59 errno, strerror(errno));
61 throw lyx::iconv_codecvt_facet_exception();
64 out_cd_ = (iconv_t)(-1);
67 virtual ~iconv_codecvt_facet()
69 if (in_cd_ != (iconv_t)(-1))
70 if (iconv_close(in_cd_) == -1) {
71 fprintf(stderr, "Error %d returned from iconv_close(in_cd_): %s\n",
72 errno, strerror(errno));
75 if (out_cd_ != (iconv_t)(-1))
76 if (iconv_close(out_cd_) == -1) {
77 fprintf(stderr, "Error %d returned from iconv_close(out_cd_): %s\n",
78 errno, strerror(errno));
82 virtual result do_out(state_type &, intern_type const * from,
83 intern_type const * from_end, intern_type const *& from_next,
84 extern_type * to, extern_type * to_end,
85 extern_type *& to_next) const
87 size_t inbytesleft = (from_end - from) * sizeof(intern_type);
88 size_t outbytesleft = (to_end - to) * sizeof(extern_type);
91 result const retval = do_iconv(out_cd_,
92 reinterpret_cast<char const **>(&from_next),
93 &inbytesleft, &to_next, &outbytesleft);
94 if (retval == base::error) {
96 "Error %d returned from iconv when converting from %s to %s: %s\n",
97 errno, ucs4_codeset, encoding_.c_str(),
99 fputs("Converted input:", stderr);
100 for (intern_type const * i = from; i < from_next; ++i) {
101 unsigned int const c = *i;
102 fprintf(stderr, " 0x%04x", c);
104 unsigned int const c = *from_next;
105 fprintf(stderr, "\nStopped at: 0x%04x\n", c);
106 fputs("Unconverted input:", stderr);
107 for (intern_type const * i = from_next + 1; i < from_end; ++i) {
108 unsigned int const c = *i;
109 fprintf(stderr, " 0x%04x", c);
111 fputs("\nConverted output:", stderr);
112 for (extern_type const * i = to; i < to_next; ++i) {
113 // extern_type may be signed, avoid output of
114 // something like 0xffffffc2
115 unsigned int const c =
116 *reinterpret_cast<unsigned char const *>(i);
117 fprintf(stderr, " 0x%02x", c);
124 virtual result do_unshift(state_type &, extern_type * to,
125 extern_type *, extern_type *& to_next) const
127 // utf8 does not use shifting
131 virtual result do_in(state_type &,
132 extern_type const * from, extern_type const * from_end,
133 extern_type const *& from_next,
134 intern_type * to, intern_type * to_end,
135 intern_type *& to_next) const
137 size_t inbytesleft = (from_end - from) * sizeof(extern_type);
138 size_t outbytesleft = (to_end - to) * sizeof(intern_type);
141 result const retval = do_iconv(in_cd_, &from_next, &inbytesleft,
142 reinterpret_cast<char **>(&to_next),
144 if (retval == base::error) {
146 "Error %d returned from iconv when converting from %s to %s: %s\n",
147 errno, encoding_.c_str(), ucs4_codeset,
149 fputs("Converted input:", stderr);
150 for (extern_type const * i = from; i < from_next; ++i) {
151 // extern_type may be signed, avoid output of
152 // something like 0xffffffc2
153 unsigned int const c =
154 *reinterpret_cast<unsigned char const *>(i);
155 fprintf(stderr, " 0x%02x", c);
157 unsigned int const c =
158 *reinterpret_cast<unsigned char const *>(from_next);
159 fprintf(stderr, "\nStopped at: 0x%02x\n", c);
160 fputs("Unconverted input:", stderr);
161 for (extern_type const * i = from_next + 1; i < from_end; ++i) {
162 unsigned int const c =
163 *reinterpret_cast<unsigned char const *>(i);
164 fprintf(stderr, " 0x%02x", c);
166 fputs("\nConverted output:", stderr);
167 for (intern_type const * i = to; i < to_next; ++i) {
168 unsigned int const c = *i;
169 fprintf(stderr, " 0x%02x", c);
176 virtual int do_encoding() const throw()
180 virtual bool do_always_noconv() const throw()
184 virtual int do_length(state_type & /*state*/, extern_type const * from,
185 extern_type const * end, size_t max) const
187 // The docs are a bit unclear about this method.
188 // It seems that we should calculate the actual length of the
189 // converted sequence, but that would not make sense, since
190 // once could just do the conversion directly.
191 // Therefore we just return the number of unconverted
192 // characters, since that is the best guess we can do.
194 intern_type * to = new intern_type[max];
195 intern_type * to_end = to + max;
196 intern_type * to_next = to;
197 extern_type const * from_next = from;
198 do_in(state, from, end, from_next, to, to_end, to_next);
202 size_t const length = end - from;
203 return min(length, max);
206 virtual int do_max_length() const throw()
208 // FIXME: this information should be transferred to lib/encodings
209 // UTF8 uses at most 4 bytes to represent one UCS4 code point
210 // (see RFC 3629). RFC 2279 specifies 6 bytes, but that
211 // information is outdated, and RFC 2279 has been superseded by
213 // The CJK encodings use (different) multibyte representation as well.
214 // All other encodings encode one UCS4 code point in one byte
215 // (and can therefore only encode a subset of UCS4)
216 // Note that BIG5 and SJIS do not work with LaTeX (see lib/encodings).
217 // Furthermore, all encodings that use shifting (like SJIS) do not work with
218 // iconv_codecvt_facet.
219 if (encoding_ == "UTF-8" ||
221 encoding_ == "EUC-TW")
223 else if (encoding_ == "EUC-JP")
225 else if (encoding_ == "BIG5" ||
226 encoding_ == "EUC-KR" ||
227 encoding_ == "EUC-CN" ||
228 encoding_ == "SJIS" ||
229 encoding_ == "GBK" ||
236 /// Do the actual conversion. The interface is equivalent to that of
237 /// iconv() (but const correct).
238 inline base::result do_iconv(iconv_t cd, char const ** from,
239 size_t * inbytesleft, char ** to, size_t * outbytesleft) const
241 char const * const to_start = *to;
242 size_t converted = iconv(cd, const_cast<char ICONV_CONST **>(from),
243 inbytesleft, to, outbytesleft);
244 if (converted == (size_t)(-1)) {
248 return base::partial;
260 /// The narrow encoding
270 void setEncoding(Ios & ios, string const & encoding, ios_base::openmode mode)
272 // We must imbue the stream before openening the file
274 locale locale(global, new iconv_codecvt_facet(encoding, mode));
279 const char * iconv_codecvt_facet_exception::what() const throw()
281 return "iconv problem in iconv_codecvt_facet initialization";
285 idocfstream::idocfstream(string const & encoding) : base()
287 setEncoding(*this, encoding, in);
291 idocfstream::idocfstream(const char* s, ios_base::openmode mode,
292 string const & encoding)
295 setEncoding(*this, encoding, in);
300 odocfstream::odocfstream(): base()
302 setEncoding(*this, "UTF-8", out);
306 odocfstream::odocfstream(const char* s, ios_base::openmode mode,
307 string const & encoding)
310 setEncoding(*this, encoding, out);
315 void odocfstream::reset(string const & encoding)
317 setEncoding(*this, encoding, out);
322 SetEnc setEncoding(string const & encoding)
324 return SetEnc(encoding);
328 odocstream & operator<<(odocstream & os, SetEnc e)
330 if (has_facet<iconv_codecvt_facet>(os.rdbuf()->getloc())) {
331 // This stream must be a file stream, since we never imbue
332 // any other stream with a locale having a iconv_codecvt_facet.
333 // Flush the stream so that all pending output is written
334 // with the old encoding.
336 locale locale(os.rdbuf()->getloc(),
337 new iconv_codecvt_facet(e.encoding, ios_base::out));
338 // FIXME Does changing the codecvt facet of an open file
339 // stream always work? It does with gcc 4.1, but I have read
340 // somewhere that it does not with MSVC.
341 // What does the standard say?
348 #if ! defined(USE_WCHAR_T)
349 odocstream & operator<<(odocstream & os, char c)
358 #if ! defined(USE_WCHAR_T) && defined(__GNUC__)
359 // We get undefined references to these virtual methods. This looks like
360 // a bug in gcc. The implementation here does not do anything useful, since
361 // it is overriden in iconv_codecvt_facet.
364 template<> codecvt<lyx::char_type, char, mbstate_t>::result
365 codecvt<lyx::char_type, char, mbstate_t>::do_out(
366 mbstate_t &, const lyx::char_type *, const lyx::char_type *,
367 const lyx::char_type *&, char *, char *, char *&) const
373 template<> codecvt<lyx::char_type, char, mbstate_t>::result
374 codecvt<lyx::char_type, char, mbstate_t>::do_unshift(
375 mbstate_t &, char *, char *, char *&) const
381 template<> codecvt<lyx::char_type, char, mbstate_t>::result
382 codecvt<lyx::char_type, char, mbstate_t>::do_in(
383 mbstate_t &, const char *, const char *, const char *&,
384 lyx::char_type *, lyx::char_type *, lyx::char_type *&) const
391 int codecvt<lyx::char_type, char, mbstate_t>::do_encoding() const throw()
398 bool codecvt<lyx::char_type, char, mbstate_t>::do_always_noconv() const throw()
403 #if __GNUC__ == 3 && __GNUC_MINOR__ < 4
406 int codecvt<lyx::char_type, char, mbstate_t>::do_length(
407 mbstate_t const &, const char *, const char *, size_t) const
415 int codecvt<lyx::char_type, char, mbstate_t>::do_length(
416 mbstate_t &, const char *, const char *, size_t) const
424 int codecvt<lyx::char_type, char, mbstate_t>::do_max_length() const throw()